MCP-Repo2LLM
# MCP-Repo2LLM
## Overview
mcp-repo2llm is a MCP server that transforms code repositories into LLM-friendly formats.
A powerful tool that transforms code repositories into LLM-friendly formats, bridging the gap between traditional code bases and modern AI language models.
This repo is based on [RepoToTextForLLMs](https://github.com/crisschan/RepoToTextForLLMs), which provides core functionality for converting repositories into LLM-readable formats.
## Motivation
As AI and Large Language Models (LLMs) become increasingly important in software development, there's a growing need to effectively communicate our codebases to these models. Traditional code repositories aren't optimized for LLM processing, which can lead to suboptimal results when using AI tools for code analysis and generation.
## Problem Solved
This project addresses several critical challenges:
- Difficulty in processing large codebases with LLMs
- Loss of context and structure when feeding code to AI models
- Inefficient handling of repository metadata and documentation
- Inconsistent formatting across different programming languages
## Key Features
- **Smart Repository Scanning**: Intelligently processes entire codebases while maintaining structural integrity
- **Context Preservation**: Maintains important contextual information and relationships between code files
- **Multi-language Support**: Handles various programming languages with language-specific optimizations
- **Metadata Enhancement**: Enriches code with relevant metadata for better LLM understanding
- **Efficient Processing**: Optimized for handling large repositories with minimal resource usage
## Installation
To install mcp-repo2llm by uv:
```
"mcp-repo2llm-server": {
"command": "uv",
"args": [
"run",
"--with",
"mcp[cli]",
"--with-editable",
"/mcp-repo2llm",
"mcp",
"run",
"/mcp-repo2llm/mcp-repo2llm-server.py"
],
"env":{
"GITHUB_TOKEN":"your-github-token",
"GITLAB_TOKEN":"your-gitlab-token"
}
}
```
GITHUB_TOKEN: your github token
GITLAB_TOKEN: your gitlab token
## Tools
### get_gitlab_repo
- Process and return the code from a GitLab repository branch as text
- Input:
- repo_url (string): the repository URL from gitlab
- branch (string): The branch name,default is master
- Returns(string): The project all information and struction from the repository as text
### get_github_repo
- Process and return the code from a Github repository branch as text
- Input:
- repo_url (string): the repository URL from github
- branch (string): The branch name,default is master
- Returns(string): The project all information and struction from the repository as text
### get_local_repo
- Process and return the code from a GitLab repository branch as text
- Input:
- repo_url (string): the repository path
- Returns(string): The project all information and struction from the repository as text
TDQS
Scored across 3 tools
Each tool has a clearly distinct purpose targeting a different repository source: GitHub, GitLab, and local repositories. The descriptions explicitly differentiate them by source, leaving no ambiguity about which tool to use for each scenario.
All tools follow a consistent verb_noun pattern with 'get_' prefix and descriptive suffixes indicating the repository source. The naming is uniform and predictable across all three tools.
With only 3 tools, the server feels thin for its apparent scope of repository code retrieval. While it covers three repository sources, the functionality is limited to a single 'get' operation without supporting related actions like searching, filtering, or managing repositories.
The tool surface is severely incomplete for repository interaction. It only provides retrieval operations without any create, update, delete, search, or management capabilities. Even within retrieval, there are no options for filtering, pagination, or handling different file types beyond returning 'code as text'.